10.7.3
Cashflow Ratios
Cashflow ratios determine how much cash the company is generating from their sales and the amount of cash they have to cover obligations. Cash flow ratios are more reliable indicators of liquidity than balance sheet or income statement ratios such as the quick ratio or the current ratio. Sometimes a company that appear very profitable can actually be at a financial risk if they are generating little cash from these profits. There have been many studies on cash flow and cash flow related variables for bankruptcy prediction. Jooste (2007) determined that bankrupt companies have lower cash flows than non-bankrupt companies furthermore findings were that in- come statement and balance sheet ratios were not enough to measure liquidity. A company can have positive liquidity ratios and increasing profits, yet have serious cash flow problems.
10.7.4
Solvency Ratios
Solvency ratios, also referred to as leverage ratios, indicate a company’s financial health in the context of its debt obligations. They identify going concern issues and a firm’s ability to pay its debt in the long term.
10.8
Description of Variables in this study
The purpose of this study as seen in Chapter 1 is to compare the performance of Machine Learning Techniques, in particular support vector machines and neural net- works, to traditional statistical models in a south african context. This study uses a set of ratios selected to evaluate the financial performance of each company. The selection and the choice of ratios were based on three main conditions:
• The ratios have been frequently used in past studies
• The ratios have been shown to perform well in past studies
• The ratios represent liquidity ratios, profitability ratios, cash flow ratios and solvency ratios.
In order to select these variables a matrix of studies was created in Microsoft Excel. A simple count function is used to find the popular ratios used. This can be seen
94 CHAPTER 10. FEATURE SELECTION
in Figure 10.3. 23 financial ratios were chosen. These 23 ratios comprise of those representing liquidity ratios, profitability ratios, cash flow ratios and solvency ratios as can be seen in Figure 10.4.
10.8. DESCRIPTION OF VARIABLES IN THIS STUDY 95
96 CHAPTER 10. FEATURE SELECTION
10.8. DESCRIPTION OF VARIABLES IN THIS STUDY 97 A large number of authors have referred to the literature to select their final variables. This is the case for Deakin (1972) with Beaver (1966) model. This method of selecting predictors may be relevant when the aim is to look into the conditions of replicating existing models but such a strategy is not efficient for at least two reasons. First, the performance of a variable is not stable. Second, the predictive ability of one variable cannot be assessed in isolation, but in conjunction with others and with a specific modeling technique. A good variable or set of variables does not exist in itself; a good set of variables seems to be in part the result of the characteristics of the set itself and that of the fit between this set and the modeling method. Therefore choosing bankruptcy predictors solely for their popularity in the literature is not sufficient (du Jardin P, 2012).
du Jardin P (2012) also states that relying on statistics and selecting features on the basis of their scores in various statistical tests (variable ranking) has also proven to be insufficient.
Guyon (2003) outlines the limitation of variable ranking techniques and presents several situations in which the variable dependencies cannot be ignored. He points out that noise reduction and better class separation may be obtained by adding variables that are presumably redundant. He also stated that individual variables can have no separation power by themselves but taken together, the variables can provide good class separability. Guyon (2003) points out that even though perfectly correlated variables are truly redundant in the sense that no additional information is gained by adding them, very high variable correlation does not mean absence of variable complementarity. It was on this basis that a wrapper selection technique was used for variable selection rather than preforming statistical analysis for feature selection.
This study chooses a set of dominant ratios derived from a larger set of related ratios as a prior screening tool or pre processing step. There are hundreds of financial ratios that can be constructed from a set of financial statements and thus preforming an analysis of this type on all of them would be too exhaustive. These ratios were selected to evaluate liquidity, profitability, cash flow and solvency. After these 23 Ratios were selected a feature selection was used to choose the subset of variables in order to improve classification accuracy. A forward elimination wrapper method was used.
This is similar to the technique used by (Muller, 2008). Muller (2008) who first used the input variables from previous studies, followed by a feature selection technique for choosing the optimal input variables. Wu et al. (2006) also employed variables successful in bankruptcy prediction modelling as a screening tool to create a subset of variables after which a more formal evaluation was made.
98 CHAPTER 10. FEATURE SELECTION